Published

Technology and Scienceupdated 20m ago2 min read

Google GTIG Warns exploit_agent Automates Vulnerability Research and Exploits From Patches

16outlets comparedWhere the coverage agrees

10 Other3 Western Alternative1 Asian2 Other source types

This article is an automated aggregation of coverage from news outlets. Our editorials offer in-depth analysis curated by the NewsCord team.

Threat actors use AI agents to identify patches, discover flaws, and weaponize exploits without humans. Exploit_agent frameworks compare patches, draft exploits, and launch exploits with no human in loop.

Sources

16 outlets in this story. Read their headlines below.

Patch to exploit, unaided

Google Threat Intelligence Group said it identified exploit_agent, a multi-agent framework that automated vulnerability research and exploitation without a human in the loop. Google said the system could continuously scan software updates, compare code before and after patches, and identify fixed vulnerabilities. Google said the framework could then plan targeted attacks at organisations that had not updated to the new version.

Google said it found the framework on an exposed staging server connected to several frontier and locally hosted LLMs, and it also contained data taken from victims and scripts for internal reconnaissance, brute-force attacks, installing webshells and removing webshells deployed by rival attackers. GTIG said it warned that machine-speed threats could not be countered at human speed, and it urged companies to embrace agentic defence.

Reporting for this sectionIndia Today

From speed to scale

John Hultquist, chief analyst at GTIG, said attackers were still going after known, unfixed issues, “just with much greater speed and scale,” and he argued that companies needed to “embrace agentic defence.” Google said the shift marked a move from hackers using AI for individual tasks to deploying it across automated attack workflows, including exploitation pipelines against internet-facing servers.

The Google Threat Intelligence Group report said vulnerability disclosures “more than doubled between January and August,” while exploited zero-day vulnerabilities averaged eight per month in 2025 and 11 per month from January through August 2026, reaching 22 in August.

The report cited an August 2026 PaperCut campaign in which attackers achieved domain administrator control within minutes, and it said GreyNoise attributed that speed to the automated workflow.

Cloudflare also reported attempts to exploit the PHP flaws against its customers in October, and it said the two PHP vulnerabilities involved were discovered using LLM analysis after technical information published before the Oct. 21 patch gave researchers and attackers material to study.

Reporting for this sectionIndia TodayCointelegraph

What defenders face next

Google said the exploit_agent case showed how automated workflows reduce the time defenders have to deploy security patches before attackers turn disclosed weaknesses into usable exploits. Google said the same report warned that automated tools can produce false leads, and it pointed to the PHP2Shell example where hundreds of purported proof-of-concept exploits, mostly AI-generated, appeared shortly after PHP announced two flaws. Google said it projected that threat actors could launch unsophisticated attacks requiring “little to no human interaction” within the next three months, and it said “sophisticated, fully autonomous” agentic attacks were likely to emerge in the next six months.

Google said it had observed financially motivated and state-sponsored threat groups using agentic workflows powered by AI-based offensive frameworks, including CAIRN in an attack against the U.S. aerospace sector and suspected North Korean hackers using Claude Code to deliver malware. Google said it disabled accounts associated with misuse of its services and used the findings to improve safeguards, while it urged companies to check which applications need updating and apply security fixes promptly, especially on systems accessible from the internet.

Reporting for this sectionBleepingComputerHackread